Implement `read` builtin about moor

[rdaum]

In particular the problem here is how this fits with the I/O transaction model. Or doesn't. It's possible this one isn't going to happen, and will require core modifications to avoid.

[rdaum]

Thinking about this some more, it's really kind of a variant of suspend with the caveat that it needs the network host to take some action.

So two two things:

1. Implement the flow on Session and back through the RPC layer: It should be a fork on the Session instance (just like suspend()) but followed by a channel event sent back to the client, requesting the input. And then the host should do the right thing to pull the input from the client.
2. Transactional semantics, which are really the same as for suspend(); commit the current transaction ("world state") and start a new one.

[rdaum]

In any case the big caveat here is that just like with suspend, all bets are off on state / transactional consistency pre- and post- the read; there should be a big "buyer beware" message on both of those for verbs that use them, because they both introduce the potential for race conditions or messy state.

[rdaum]

Implemented in 813e0cb

As mentioned above: the read() invocation commits the current transaction, flushes the current I/O session buffer, and starts a new transaction for the post-read scenario.